1. Field of the Invention
The present invention relates to an image analysis method, a camera apparatus, a control apparatus, a control method and a storage medium.
2. Description of the Related Art
A conventional technique for detecting a specific target such as a face or a human body from an image captured by a surveillance camera ordinarily uses one matching pattern (dictionary) in which features of a target object (object) are stored, or a plurality of prepared matching patterns at different angles. Then, the object that matches such a matching pattern is detected from a detection target region in the image.
For example, Japanese Patent Laid-Open No. 2009-211311 discloses a technique in which a plurality of local image patterns are extracted using a matching pattern as the entire image by changing the size of the image, and a local feature amount is calculated from each local image pattern. Then, a determination is made as to whether or not the image includes a human body, from a total sum of the results obtained by weighting the local feature amounts.
Meanwhile, a system such as a surveillance system is known that includes a plurality of cameras, and tracks and monitors a moving object by using the plurality of cameras in cooperation.
Furthermore, US-2005 -0057653 discloses a method for continuously tracking a moving body by extracting feature information of the moving body through image recognition while tracking and monitoring the moving body and transmitting the feature information between cameras via a network.
Under the circumstances in which network camera devices have an increased number of pixels, and in which highly accurate real-time processing is required to be performed in monitoring or the like using network camera devices, detection processing for detecting a specific object is required to be executed at a high speed.
For this reason, various methods have been proposed such as reducing the size of a processing region by selecting only a region in which there is a movement from an image as a target region area, or reducing the frequency at which detection processing is performed to a processable level by thinning out processing image frames in time series.
However, the processing load of object detection processing through pattern matching or the like is high, and thus real-time processing is not always possible.
Also, in images captured by a surveillance camera, a target subject moves in various directions and thus may take postures in various orientations. Accordingly, the processing load further increases when performing highly accurate pattern matching by using matching patterns (dictionary) regarding a plurality of orientations such as a front orientation and a side orientation.
As described above, under the current circumstances, there is a problem in that there is a trade-off relationship because in order to perform object detection processing with high accuracy, the processing load increases and the processing speed decreases, whereas in order to perform object detection processing at a high speed, the use of a low-load detection unit is required.